First Monday

Zooming into an Instagram City:
Reading the local through social media by Nadav Hochman and Lev Manovich

How are users’ experiences of production, sharing, and interaction with the media they create mediated by the interfaces of particular social media platforms? How can we use computational analysis and visualizations of the content of visual social media (e.g., user photos, as opposed to upload dates, locations, tags and other metadata) to study social and cultural patterns? How can we visualize this media on multiple spatial and temporal scales? In this paper, we examine these questions through the analysis of the popular mobile photo–sharing application Instagram. First, we analyze the affordances provided by the Instagram interface and the ways this interface and the application’s tools structure users’ understanding and use of the “Instagram medium.” Next, we compare the visual signatures of 13 different global cities using 2.3 million Instagram photos from these cities. Finally, we use spatio–temporal visualizations of over 200,000 Instagram photos uploaded in Tel Aviv, Israel over three months to show how they can offer social, cultural and political insights about people’s activities in particular locations and time periods.


(not) On photography
Function within relation
Social TimeSpace
Data visualization and imaginary communities




A large proportion of contemporary cultural media is created, experienced, and shared using software. Contrary to earlier incarnations of the web that were focused on content created by professionals, companies, and organizations, we are now producing, sharing, or tagging massive amounts of our own images and videos. If, similar to earlier technological developments, this media shift changes the way we know the world, and the ways in which we generate and conceive this knowledge, what can media sharing software tell us about our own experience of living in the present? How can we trace the socio–cultural operations of this software as well as its imaginations and potentials? What can visual social media tell us about the lives of cities, neighborhoods, and individuals? What is this data not able to reflect, and what can it only show with systematic distortions?


Radial image plot visualization of 33,292 photos uploaded to Instagram in Tel Aviv during 20-26 April 2012
Figure 1: (a) Radial image plot visualization of 33,292 photos uploaded to Instagram in Tel Aviv during 20–26 April 2012. The photos are sorted by hue (radius) and upload time (perimeter). (b) Radial image plot visualization of a random sample of 50,000 photos uploaded to Instagram in Tel Aviv. The photos are organized by brightness median (perimeter) and hue median (radius). Higher resolution versions of these figures are available at High–resolution versions of the visualizations in this article are available at


This paper addresses these questions via the analysis of the mobile photo–sharing application Instagram, a social network that offers its users a way to upload photos, apply different manipulation tools (‘filters’) in order to transform the appearance of an image, and share them instantly with the user’s friends (using Instagram’s application or other social networking sites such as Facebook, Foursquare, Twitter, etc.) [1]. As of June 2013, only three years after its launch, the application already has over 130 million registered users who have shared nearly sixteen billion photos from all over the globe [2].

In the first part of this article we analyze Instagram’s core features and examine some of the ways in which users interact with the application. We then use “media visualization” techniques developed in our lab to explore visual patterns across large sets of Instagram photos. We start with 2,353,017 Instagram photos from 13 cities around the world. We then look in more detail at 212,242 photos uploaded by users in Tel Aviv, Israel over a three–month period. Next, we examine photos taken over two weeks during national events in this city, and finally focus on individual days during this period.

The goal of this exploration is to show how globally shared media and metadata can be used to study patterns on multiple scales. While many social media visualizations and computer science papers focus on large data sets aggregated in space and time (for instance, visualizations of movements of people in a city over a long period of time, or Twitter activity across the world [Fischer, 2010; Ernewein, 2013]), we suggest that social media can also be used for reading of local social and cultural events. In other words, we do not necessarily have to aggregate user generated content and digital traces for the purpose of Durkheim–like mapping of society where individual people and their particular data trajectories and media diaries become invisible. The individual and the particular do not have to be sacrificed for the sake of data aggregation, or “large scale patterns.” Instead, we can perform “thick visualization” (or, to use Todd Presner’s phrase, “thick mapping” [3]) of the data, practicing “data ethnography”, and following individuals rather than only “society.” To illustrate how this can be done in practice, we use a sample set of Instagram photos and their metadata uploaded by users in Tel Aviv during selected days corresponding to important national events. We visualize these photos in multiple ways, and demonstrate how such visualizations can lead to cultural, social, and political insights about particular local places during particular time periods.

Computer science researchers typically identify general patterns and regularities in the data, and work on models that fit this data. After the explosion of social media in 2004–2005, many researchers used this standard approach to study massive social data from Flickr, Twitter, YouTube, and other social networks (Zheng and Hong, 2012). On the other hand, digital humanities and digital history scholars so far only used computational and/or visualization techniques with sets of historical artifacts, and did not analyze contemporary social media [4]. In addition, current investigations of both historical collections and contemporary social media typically use standard information visualization techniques (i.e., bar graphs, scatter plots, timelines, network diagrams, etc.), which can only show patterns in metadata (such as image tags), and are not ideal for the exploration of visual characteristics of image sets and their content.

In contrast, our paper explores a different approach that we think is appropriate for the humanistic analysis of user–generated content and data. We use high–resolution visualizations that show complete image sets to enable the exploration of both photos’ metadata (upload dates, filters used, spatial coordinates), the patterns created by the content of their photographs, and the examination of individual photographs (see Figure 1 for an example of such visualization). We render these visualizations at the maximum size possible: 10,000 x 10,000 pixels, 20,000 x 20,000 pixels, or even larger. High–resolution versions of most visualizations appearing in this paper are available at

Such visualizations fit well with our strategy which we call “multi–scale reading.” They work equally well with massive sets of photos taken in different cities, all photos taken in a single city, or all photos shared by particular users. This ability to visualize photographic content at multiple scales allows us to start asking questions such as: How can we compare millions of photos taken in London, Bangkok and Tel Aviv in such a way that cultural differences between these cities can be revealed? Or, how can we visualize the “stories” made up by the individual users’ sequences of photos? In other words, we can study both large scale patterns and the particular unique trajectories, without sacrificing one for another.

In summary, this paper combines perspectives from social computing, digital humanities, and software studies in order to “read” and analyze visual social media data. Similar to researchers in the field of social computing, we study large sets of contemporary user generated social media, and use computational approaches in our analysis. We respond to the key question of digital humanities – how to combine “distant reading” of patterns with “close reading” of particular artifacts – by proposing a multi–scale reading. To accomplish this in practice, we use special visualization techniques (radial image plot, and image montage), which show all images in a large set organized by metadata and/or visual properties. Finally, we follow software studies paradigm by looking very closely at the interfaces, tools and affordances of the software (in this case Instagram) that enable the practice of social media.



(not) On photography

First launched in October 2010, Instagram did not seem to offer anything genuinely new compared to existing media sharing services that had similar features, such as image manipulation tools, location annotation of photos, and instant sharing. However, it is the congruent operation of these elements within a single mobile application and the presentation — i.e., how the application allowed users to create, share, and organize information — that might provide a plausible explanation for Instagram’s widespread adoption, and how it meshes with current cultural trends.

The most prominent element that underlies Instagram’s structure is its reliance on geo–temporal tagging: the geographical and temporal identification of a media artifact [5]. This is, of course, a fixed definition, but its data presentation in a specific media environment is what gives it its cultural meanings and ramifications.

For instance, Instagram’s interface suppresses temporal, vertical structures in favor of spatial connectivities. Although each image taken by the application is stamped with a specific time and place, the photos are not organized according to the Gregorian calendar but rather by a dynamic time span. The time element is always user–centric and its measurement is relative between the present moment of launching the application and the original date of creation.

This means that although the specific time in which a photo was taken exists in the software’s database, its timestamp is dynamic as each image shows a constantly changing representation of time. For example, if I currently see a photo that was taken by a friend “4 days ago”, when I open the application tomorrow the time indication will be “5 days ago”. In this way, the representation of time in relation to each image becomes elusive and remains in flux as time passes, changing from 53 seconds to 5 days, to 12 weeks, and one year ago.

While Instagram eliminates static timestamps, its interface strongly emphasizes physical place and users’ locations. The application gives a user the option to publicly share a photo’s location in two ways. Users can tag a photo to a specific venue, and then view all other photos that were taken and tagged there. If users do not choose to tag a photo to a venue, they can publically share their photos’ location information on a personal “photo–map”, displaying all photos on a zoomable world map (Figure 2).


Left to right: Instagram's timeline, filters page, and photo map
Figure 2: Left to right: Instagram’s timeline, filters page, and photo map. Source: Instagram official screenshots,, accessed 5 June 2013.


This privileging of space over time is reinforced by the organization of photos within the application. The default presentation of images does not employ groups of documented events (or private albums), which may contain each user’s photos and create a cohesive narrative. Instead, photographs are presented as a continuous stream of images from various users. Users perceive a montage of images taken by people they follow, thus eliminating notions of “traditional” time or event presentations and cataloging [6].

This notion is strengthened once again when we consider Instagram’s filter functions. While (or after) taking a photo, the application allows its users to apply different manipulation tools. By adding hues, grain, contrast, etc., each filter evokes a different “feel” changing the message communicated by an image. In this way, while taking a photo of a specific time and place, we apply a filter to it to suggest a different time or atmosphere (some of the filters are even named to suggest particular time, such as the filter called “1979”).

The result is a multi–temporal image which suggests at least three different temporal references: the actual time when the picture was taken, the time evoked by a certain filter, and the time span indicated by the application when viewing the photo. Ironically, while a geo–temporal tagged image connotes the precision of time and space coordinates (we know the exact longitude/latitude coordinates together with the exact time it was taken) the software subverts this message by displaying multiple users’ photostreams in a single feed, a relative time indication, and a distorted, filtered photographic image [7].

As a result of this distorted structure and presentation of time within the application (there is no specific time or “history” for each image) what we get is a coexistence or contemporaneous state in which all photos occur to us at the same time, no matter how different they are, when or where they were taken. In a paradoxical way, the temporal image becomes atemporal [8]. And as images become “timeless” (or better, time–thickened), we are all in the same times together.

This sense of atemporality is established not only by Instagram’s filters or time presentation, but also by its instant photo sharing function. What underlies this structure is an emerging operative cultural logic in which an individual photo is being related to a whole that potentially promises any image from any vantage point. If we follow a similar logic we can think of Instagram’s users’ extensive documentation efforts as comparable to the planetary documentation endeavors, taken, for example, by Google Earth or Bing Maps [9]. This hypothesis might also partially explain Instagram’s extensive filter usage. While Google Earth’s documentation efforts are presented as objective and detached since the service uses satellite photography, (or in the case of Google Street View, captured from specially equipped cars and stitched into continuous panoramas [10]) Instagram’s photos resonate with more personal, “authentic” experiences that chronicle the world in a way that resists the time and place represented by larger impersonal corporate documentation efforts.

These two inherently different pictorial logics can be related to an earlier similar development in visual culture: the development of Impressionism in the 1870s, shortly after the invention and spread of photography. As this historical relationship is usually described, Impressionist artists rejected the high “photographic” realism associated with the academic style, and the pursuit of visual details offered by dispassionate recordings of photographic plates. Impressionists were more concerned with the way in which the eye and intellect perceive the changing qualities of light, movements and objects. In a very similar way, Google Earth/Bing and Instagram’s two very different pictorial logics confront us with two distinct ways of seeing: an objective, elevated and fixed form versus grassroots documentation efforts that present spontaneous and highly personal sentiments that inherently reject the technological pursuit of fine details and accuracy of a “mechanical” (now digital) eye.

These two “logics” have been recently merged in certain ways, as Google now enables users to add their own geospatial data to the default Google Earth representation, creating complex and media rich projects on top of existing geographical information. In this way, users can now negate or complement the maps of Google Earth rendered in General Perspective Projection [11] by uploading new layers of geoinformation. In its most recent development, Google Earth now includes publicly sourced aerial images from balloons and kites, a grassroots mapping project in which anyone with a digital camera can attach it to a balloon or a kite and capture images that are then stitched together into a geo-referenced image [12].

The change in Google Earth’s structure into a platform for users to build upon, to compete with, to complicate and elaborate on, presents us with images that seek to pin down current changes in time, place, media (how they were taken), and mood (why they were taken). This is also the operative logic of Instagram, encouraging people to understand themselves as time and place (due to the very nature of the geo–temporal image) while offering a profound immersion in planetary documentation mechanisms. However, despite these changes, Instagram experience remains fundamentally different from that of Google Earth/Bing Maps. In the latter, images, videos, and additional data layers are secondary to the primary representation: the zoomable maps, presented as objective data. In the former, the stream of photos taken by people from a human point of view and height remains primary, with the map showing photo location delegated to a secondary function.



Function within relation

Thus far we have been tracking an operative software logic in which an individual is always being related to a documentary whole (for example, the tags and hashtags of a particular user are related to the tags and hashtags by all other users; a user’s photos are related to all other photos via a shared coordinate system). But while Instagram’s primary goal is similar to larger organizational documentation efforts, its user interface also has a secondary goal: to represent our collective visual experience differently from the ways it was represented before. Instagram signifies a new desire to creatively place together old and new — local and global — parts and wholes — in various combinations. If this is indeed true, and Instagram’s photo universe and its presentation addresses all these interests, how can we gain insights from the study of this large–scale global cultural dataset? What are the ways in which Instagram photos operate in relation to each other, and how can we trace these connections, relations and functions on a global and local scale?

1. Instagram data

Our work takes advantage of the particular characteristics of Instagram’s software. Instagram automatically adds geospatial coordinates and time stamps to all photos taken within the application. All photos have the same square format and resolution (612 x 612 pixels). Users apply Instagram filters to large proportion of photos that give them an overall defined and standardized appearance (In our sample of 2.3 million photos, the proportions of filtered photos varied between 68 and 81 percent depending on the city).

Using Instagram’s official API and the latitude and longitude data it provides, we crawled Instagram photos, and their metadata (user ID, location, comments, number of ‘likes’, date and timestamp, type of filter applied, and user–assigned tags) from 13 cities around the world. Table 1 shows the number of photos in our data set for each city, the number of unique users who uploaded these photos, and the dates for every city.


Table 1: Details of Instagram data set collected for this study
CityNumber of photosNumber
of users
DatesUsers with
> 30 photos
San Francisco344,07049,1297 Dec 2011 — 21 Apr 20124.3%
Tokyo298,48438,70411 Oct 2011 — 20 Jun 20124.7%
London236,26233,83723 Dec 2011 — 10 Apr 20124.1%
Moscow234,28923,7163 Feb 2012 — 14 Apr 20126.7%
Tel Aviv212,24215,77324 Jan 2012 — 26 Apr 201210.9%
New York245,24840,67328 Dec 2011 — 6 May 20122%
Bangkok146,27233,61227 Feb 2012 — 12 Apr 20121.6%
Sydney136,05720,41427 Oct 2011 — 16 Apr 20123.7%
Istanbul134,33813,90326 Jan 2012 — 24 Apr 20126.8%
Singapore128,50919,64227 Feb 2012 — 18 Apr 20123.7%
Paris93,13517,5556 Jan 2012 — 16 Apr 20122.5%
Berlin78,9799,73612 Feb 2012 — 27 Apr 20125.3%
Rio64,95211,36127 Jan 2012 — 26 Apr 20123.1%
Table 1: The table shows information about the Instagram data set we collected. Columns (left to right): cities, number of collected photos, number of users, collection dates, number of users who shared more than 30 photos.


It is important to note that over the course of our data collection, Instagram’s popularity increased, new features were added, and the perception of the service was also changing. For example, Facebook acquired Instagram in April 2012, and on April 12 Instagram for Android was released. During the same period competing and complementary media sharing services were also evolving. Therefore, some of our findings may refer only to a particular period in Instagram history exemplified by the sample we collected.

It should also be emphasized that in contrast to Web users, the people who were likely to use Instagram during the period of our data collection reflected a more limited demographic. According to a 2012 Pew Internet survey of users of popular social network services, 16 percent of women and 10 percent of men using the Internet were also using Instagram; among Internet users aged 18–29, 29 percent were using Instagram [13]. Thus, as a reflection of social reality or, more precisely, as a giant photograph of social reality, Instagram only captures the curated lives of some members of society and not others.

2. Related work

Social media is studied in many disciplines from many different perspectives. Since our study focuses on the analysis and visualization of large sets of Instagram photos and their geo-spatial and temporal metadata, two research areas are particularly relevant. The first area is computer science and the subfield of social computing, which explores the possibilities of algorithmic analysis of large sets of digital images created by users of popular social media services and companies such as Flickr, Picasa, Geograph or Google Street View. These studies examine text tags and geo–spatial visual data, and offer algorithms to carry out enhanced search or scene summarization in large visual corpora (Jaffe, et al., 2006; Simon, et al., 2007); trace behavioral patterns and spatial trajectories by mapping geotagged visual data (Crandall, et al., 2009; Kisilevich, et al., 2010; Kennedy and Naaman, 2008; Hays and Efros, 2008; Li, et al., 2009; Antoniou, et al., 2010; Li, et al., 2009; Vrotsou, et al., 2011); or estimate ecological phenomena using geo–temporal tagged photos (Haipeng, et al., 2012). Other studies examine the level of “attractiveness” of photos (San Pedro and Siersdorfer, 2009), or automatically locate distinctive visual elements for a certain geo–spatial area (Doersch, et al., 2012).

The second related research area (also in computer science/social computing) is analyses of spatial data in location–based applications, such as check–in data gathered from social networks such as Foursquare (for example see Cranshaw, et al., 2012). In this case, the bias which may result from the clustering of locations of Flickr photos around famous landmarks is resolved by the myriad of venues into which people check–in. Although these studies aim to depict the “true” dynamic of a city by tracing social and place proximities (where people check–in in defined areas), they typically ignore the temporal nature of the data (when people check–in). Existing studies that do take into account temporal variations of spatial situations do not consider social media data (Andrienko, et al., 2012). Some relevant work on temporal aspects of social media data includes sentiment analysis of Twitter data (Dodds, et al., 2011; Leetaru, 2012), variations of keyword use on Twitter daily patterns across geographical locations (Naaman, et al., 2012), and extracting references to “real world” activities from text–based social media data (Grinberg, et al., 2013).

The extensive spatial coverage of Instagram’s data, together with the availability of precise location and temporal information, enables us to combine these research strategies over three levels: spatial, temporal and visual. As opposed to a more standard approach in social networks research, we do not begin with predetermined problems that need to be solved or with desired applications, but rather perform an “open–ended” exploration on various levels, moving from a comparative examination of visual proximities between cities around the world to a detailed study of a specific city and its users during particular time periods. By applying a variety of visualization techniques (some developed specifically for this project), we show how the volume, spatial coordinates and visual features of Instagram photos over time can reveal local cultural and social patterns.



Social TimeSpace

Existing representations of space using social media data emphasize the fact that space does not stand on its own as a fixed entity but rather that it is a social product, bound up with specific social realities. For example, a map that compares photos taken by tourists versus photos taken by residents visualizes individual movement around a city while illustrating different experiences of a place by various social groups (Fischer, 2010; see Figure 10). In another case, a map that uses “check–in” data characterizes different areas in a city not according to municipal borders, but by a collection of individual activities and movements (Cranshaw, et al., 2012). In each of these cases, the representation of the data constructs an imaginary “social space” that is derived from the nature of the data and from the ways it is being processed and presented.

If, as previously suggested, Instagram indeed offers a particular social experience, how then does this experience construct a space? In other words, how can we grasp, visualize, and analyze the production of such a social space within the “Instagram medium” in ways that reveal its uniqueness as a cultural form?

On the one hand, we can follow Instagram’s “objective” or intended affordances and its emphasis on a sense of “presentness” in specific time and place (derived from the immediate registration and sharing options of everyday life moments). However, this presentness is complicated by the fact that Instagram photos are typically carefully curated and edited, sparsely uploaded, and are not always shared immediately (users often upload photos at a later date that were taken hours, days and sometimes even years earlier). Given these practices, can we consider individual momentary “presentness” and Instagram’s emphasis on the “now” as a key experience of this platform? And if so, how is this space different from existing social spaces offered by other forms of social media data (tweets, messages, check–ins, etc.)?

On the other hand, we can negate existing software affordances and re–introduce dimensions that are currently concealed from Instagram’s software interface. Continuing a line of thought we discussed earlier — the ways Instagram’s interface suppresses temporal, vertical structures in favor of spatial connectivities — we can bring back to forefront the temporal dimensions of our data. As opposed to maps that show social media activity aggregated over time, and in complete opposition to Instagram’s own interface, our visualizations take into account not only the spatial aggregated forms of Instagram photos, but also their temporal organization.

Influenced by Henri Lefebvre’s rhythm analysis and his temporal understanding of place and space, we introduce two types of time that exist within Instagram photos: cyclical time and linear time [14]. Cyclical time represents the diachronic order of multiple individual photographs, combined from accurate time stamps that indicate specific date and time of day of each photo. When visualized, cyclical time represents the historic process of collective social, visual production that potentially repeats itself infinitely. For example, in our temporal image montages (grids of photographs organized by their upload time) we can identify the “rhythm” of a collective social visual production (how many photos are taken in a specific time and place) and how this rhythm unfurls over time from day to night. We can then identify deviations in cyclical times, or compare different “visual rhythms” (Hochman and Schwartz, 2012) from different places (Figure 3).


Montage visualizations comparing Instagram photos shared over four consequtive days in two cities.
Figure 3: Montage visualizations comparing Instagram photos shared over four consecutive 24-hour periods in two cities. Top: 57,983 images from NYC. Bottom: 53,498 images from Tokyo. Photos are sorted by upload date and time (top to bottom, left to right). A higher resolution version of this figure is available at:


Linear time, on the other hand, is the synchronic order of all images from a particular place and time organized according to multiple visual attributes. For example, an image montage may organize all images from a specific time and place according to average brightness or average hue of each photo, thus revealing a “signature” of dominant visual preferences that might indicate a shared experience by multiple users (Figure 4).


Samples of 4,000 random Instagram photos from Bangkok (top) and Berlin (bottom)
Figure 4: 4,000 random photo samples from Bangkok (top) and Berlin (bottom). In each montage, photos are sorted by average hue (left to right, top to bottom). A higher resolution version of this figure is available at:


However, the meaning or function of such a space does not stem from the representation of each of these elements alone (the spatial organization, or the cyclical and linear times), and they cannot be examined separately [15]. It is only by the integration of the spatial, the cyclical, and the linear that we can actually measure the production and examine the function of a social timespace: the representation of space through social media data according to its spatial and temporal organization (see for example Figure 5). Social time and space — as the combination of the cyclical and the linear times in our visualizations — are not only relational (linear) but also historical (cyclical). Our visualized social timespace is thus a representation of an active web of affinities that is constantly shaped and reshaped by users.


A radial plot visualization of 23,581 photos from the Brooklyn area during Hurricane Sandy
Figure 5: A radial plot visualization showing 23,581 photos uploaded to Instagram in Brooklyn area during Hurricane Sandy (29–30 November 2012). Photo’s distance from the center (radius) corresponds to its mean hue; photo’s angle (i.e. the position along the perimeter) corresponds to its time stamp. Note the demarcation line that reveals the moment of a power outage in the area and indicates the intensity of the shared experience (dramatic decrease in the number of photos, and their darker colors to the right of the line). A higher resolution version of this figure is available at:


Ironically, this representation – derived from the combination of the cyclical and the linear times – is not available in the application itself. Instagram’s affordances blur specific time indications and enforce uniform appearances on its photos, thus creating a sense of atemporality and shared aesthetics. Our analysis shows how Instagram’s interface superimposes its strong “message” (or “interface signature”) on its users, shaping what and how they communicate. For example, Figure 6 compares filter use across six cities in our data set and shows how the proportions between photos with different filters are remarkably similar for all cities.


Filters usage percentage by city
Figure 6: The use of Instagram filters in six cities. The filter names appear on the perimeter. Additional radial plot visualizations illustrating filter use are available at:


However, when examined on a large scale we can see that social timespace is not universal. As apposed to Instagram’s interface uniformity imposed on all application users in all places – in terms of time representation, photo dimensions, same set of filters etc. – we found small but systematic visual differences between photos shared on Instagram in different cities [16].

To study differences between cities, we first selected random samples of 50,000 photos from our larger photo sets from various cities, and extracted a number of visual features from these photos [17]. The features include basic statistics (mean, median, standard deviation, histograms, etc.) for brightness, hue, and saturation, number of edges, contrast, and texture measurements. We created radial plot visualizations which show a 50,0000 image samples from different cities organized by some of these features. For example, in Figure 7 we compare NYC and Bangkok images orgaznied by brightness mean (radius) and hue mean (perimeter) as well as San Francisco and Tokyo images organized by hue median (radius) and brightness mean (perimeter).


Figure 7: Radial plot visualizations of 50,0000 image samples organized by visual attributes. Top left: San Francisco – brightness mean (radius) and hue mean (perimeter). Top right: Tokyo — brightness mean (radius) and hue mean (perimeter). Bottom left: NYC — hue median (radius) and brightness mean (perimeter). Bottom right: Bangkok — hue median (radius) and brightness mean (perimeter). Higher resolution versions of these visualizations are available at:


Next, we selected random samples of 4,000 photos for each of the 13 cities in our data set, and similarly extracted a number of features for all photos in every city. Figure 8 shows the results of multidimensional scaling (MDS) with two different sets of these features. One set contains only nine color features; the other set ads brightness and texture measurements (16 features total). While the details differ depending on which features are used, the overall pattern is the same: Bangkok, Singapore and Tokyo are situated apart from other cities. Within the cluster formed by the remaining cities, each also occupies a different position.


MDS (multidimensional scaling) of 4,000 random samples of Instagram photos from 13 cities, a) MDS of 16 visual features
MDS (multidimensional scaling) of 4,000 random samples of Instagram photos from 13 cities, b) MDS of nine color features
Figure 8: MDS (multidimensional scaling) using selected visual features for 4,000 random samples of Instagram photos from 13 cities. (a) MDS of 16 visual features including color; (b) MDS of nine color features only. Notice that while the results depend on the visual features being used in each case, in both cases we see the same pattern: Bangkok, Singapore and Tokyo are situated apart from the rest of the cities.

This analysis of visual features of large photo samples suggests that within Instagram’s global shared photo universe, each city has a distinct “visual signature.” Thus, if Instagram’s affordances indeed offer a new global style, its universality possesses distinctive characteristics in different social timespaces. As our visualizations illustrate, to various degrees of different visual measures, all of our cities exhibit at the same time local, regional and universal character.

These alternative strategies (Figures 3, 4, 6, 7, 8) illustrate just some of the ways to compare “Instagram Cities.” But how can we explore our data on a smaller scale, to better see the formation of social timespace and its multiple modalities (spatial, cyclical, linear, and temporal)? In other words, how can the active process of production of a social timespace be visualized and analyzed? This will be addressed in the next section.



Data visualization and imaginary communities

The recent proliferation of visualization techniques (which show locations, check–ins, routes, and other social media and physical information) aggregate large amounts of data into a single condensed representation of a city, country or the Earth (Figure 9). These condensed representations usually neglect the specificity of the particular images, check–ins, and other details; privileging instead an aggregation of countless other similar forms. Most often, they do not represent a whole that emerges in specific times, but rather a whole that exists outside of time — a representational form that tells us something about the nature of a place but which rarely has the power to explain the nature of the specific time when these aggregated actions occurred. Involuntarily, they construct “imaginary communities” — visions of the whole that do not actually exist.


A real-time Twitter visualization that maps tweets across the world
Figure 9: A screenshot from a real-time visualization by Franck Ernewein showing Twitter activity around the world., accessed 5 June 2013.


These imaginary communities do not trace or encapsulate real–life temporal changes. For instance, a visualization made up of routes of millions of people aggregated over months or years creates a convincing map of a city, with its major streets alight. But this “city” does not exist, because the individual traces that compose it do not temporally coexist. These traces do not correspond to any social reality actually experienced by people. As we move through a city, we do not see traces made by other people in earlier times, we do not even see our own trajectory, and others do not see our paths (Figure 10).


A comparison of locals versus tourists' movement in the city
Figure 10: Eric Fischer (2010), “Locals and Tourists #2 (GTWA #1): New York.” The visualization compares locations of photos uploaded to Flickr and Picasa. Blue pictures are by locals. Red pictures are by tourists. Yellow pictures might be by either., accessed 5 June 2013.


Our visualizations, which display locations of Instagram photos taken by individuals over time, illustrate exactly that. When aggregated into a single visualization, an image of a city emerges (Figure 11). This image constructs an illusion of many people congregating in particular places at the same time (as captured by their Instagram actions), but in reality, most users have taken only a few photos over a specific time period and these are widely distributed in time and space. Even if we only look at several avid users, their time/space coordinates almost never intersect (Figure 12). How are we then to better trace, characterize and visualize the multitudes of users’ trajectories and photos, each following its own pattern?


Plot of location metadata for Tel Aviv photos over a three-month period
Figure 11: Locations of photos shared on Instagram in Tel Aviv over a three month period (24 January—26 April 2012). 212,242 photos were shared by 15,773 different users. The points are colored using a green to red gradient (green — morning, yellow — afternoon, red — evening). A higher resolution version of this figure is available at:



A screenshot of users' photos between 14:00pm and 14:05pm over a map of Tel Aviv
Figure 12: In order to examine space and time trajectories of most active Instagram users in Tel Aviv, we developed an interactive Web application. This application screenshot shows information about users and their shared photos during a 5 minute period. It illustrates that even the most active users rarely share photos at the same place at the same time.


1. Collective memory routines

A possible thread to follow is to look at exceptional times in specific places. We chose to examine, what are arguably, three of the most emotionally, culturally, and politically charged days in Israeli society, and the ways in which these days were experienced in Tel Aviv in 2012: Holocaust and Heroism Remembrance Day (18–19 April 2012), Israeli Fallen Soldiers and Victims of Terrorism Remembrance Day (24–25 April), and Israeli Independence Day (25–26 April).

We start with a comparison of the two memorial days. On both days, most events begin at sunset and include numerous ceremonies on the national and regional levels, with countless services performed in schools, city centers, and cultural hubs. Both days include sirens that sound throughout the country for a few minutes at a time, with people standing still while remembering the dead (a two–minute siren during the Holocaust day on 10:00am, and two sirens during the Fallen Soldiers’ Day, one at 8:00pm and the other one at 11:00am the next day).

The Israeli Memorial Day is significantly different than those in other countries such as Memorial Day in the U.S. For example, on Isreal’s Memorial eve, places of public entertainment are legally closed. During these days, all cable channels go dark, Israeli television channels only air special documentaries about war victims and the fallen soldiers, and solemn songs are played on the radio (Figure 13).

What can we learn from Instagram’s data about the structures of these emotionally charged days? Can we see differences in the ways these days are treated in contemporary Israeli society? What type of insights can we extract from the ways individual users choose to spend their remembrance days? Or in other words, what kind of stories do their “photo trails” tell us about the nature of these days and their cultural significance?


Israelis stand still during a two-minute siren for Soldiers Memorial Day
Figure 13: Israelis stand still during a two–minute siren for Soldiers Memorial Day. Source: Dan Bar Dov (2008),, accessed 5 March 2013.


2. The Spread of Sorrow

Although both of these memorial days are of similar cultural magnitude and are accompanied by similar ceremonial routines, our results illustrate behavioral differences in the way Instagram users perceive and experience them. Interestingly, we did not find very significant differences between the Holocaust Memorial Day and other days during that week. As for the Fallen Soldiers Memorial Day, however, our data shows significant differences from regular daily patterns on every dimension: geospatial coverage (spatial distribution of the locations where photos were taken), the volumes of photos being shared, and their content. Below we discuss our findings in more detail.

Holocaust and Heroism Remembrance Day

Given the historical and emotional significance of the national and private memorial routines observed during Holocaust Memorial Day eve (and the following day) all around the country, one might expect to see deviations in photo–taking habits compared with regular days. However, as our results show, Tel Aviv Instagram users remain indifferent overall and do not share a significantly different number of photos during Memorial Day than on any other day (Figure 14a).

Holocaust Memorial Day eve (18 April) was accompanied by a slight decline in the number of photos (25 percent less than the average amount of photos in the previous three evenings, between 8pm and midnight). This might be explained by the fact that many entertainment venues and businesses such as bars and restaurants are closed during the evening. During Holocaust Memorial Day (19 April), however, Instagram photo–sharing patterns remain similar to other days. We do note an unusual decline in the number of pictures around 18:00pm, as well as a peak around 22:00pm. These correspond to activities that mark the end of Memorial Day and the return to everyday routines (i.e., going out, socializing, etc.) (see Figure 14c).

If we only compare Holocaust Memorial Day to other days of the week, the photo–taking volume on that day does not show notable differences. Thus, it appears at first glance that Instagram activity on that day does not reflect its national significance. However, when compared with the Israeli Fallen Soldiers Memorial Day, which takes place exactly a week after, the difference in the socio–cultural significance of these two memorial days as well as the differences between them and other days become dramatically visible.


Numbers of
photographs captured and shared on Instagram during exceptional events in the Tel Aviv area between 15-19 April and 22-26 April 2012, a) 15-19 April 2012: 17,923 photos, 5,095 users
Numbers of
photographs captured and shared on Instagram during exceptional events in the Tel Aviv area between 15-19 April and 22-26 April 2012, b) 22-26 April 2012: 23, 257 photos, 6,333 users
Numbers of
photographs captured and shared on Instagram during exceptional events in the Tel Aviv area between 15-19 April and 22-26 April 2012, c) Holocaust Memorial Eve and Day, 18-19 April 2012: 7,055 photos, 2,993 users
Numbers of
photographs captured and shared on Instagram during exceptional events in the Tel Aviv area between 15-19 April and 22-26 April 2012, d) Israeli Fallen Soldiers and Victims of Terrorism Remembrance Day, 24-25 April 2012: 8,631 photos, 3,519 users
Figure 14: Numbers of photographs captured and shared on Instagram during exceptional events in the Tel Aviv area between 15–19 April and 22–26 April 2012. (a) 15–19 April 2012: 17,923 photos, 5,095 users. (b) 22–26 April 2012: 23, 257 photos, 6,333 users. (c) Holocaust Memorial Eve and Day, 18–19 April 2012: 7,055 photos, 2,993 users. (d) Israeli Fallen Soldiers and Victims of Terrorism Remembrance Day, 24–25 April 2012: 8,631 photos, 3,519 users. Red Bars indicate decrease in number of pictures taken (see text for discussion).


Israeli Fallen Soldiers and Victims of Terrorism Remembrance Day

In the second Memorial Day (24–25 April), we see a significant decrease in the numbers of shared photographs after the siren is sounded across the country. Between 20:00pm and 21:00pm, 50 percent fewer photos were uploaded when compared with the average number of photos in the same time period during the previous five days (see Figure 14b). When the second siren sounded the next morning, the volume of shared photos increased due to the many ceremonies taking place around the city immediately after (Figure 14d).

Although both memorial days play a similar role in national memorial practices, our data reveals significant deviations between them. While behavioral patterns during Holocaust Memorial Day do not show exceptional deviations from regular daily patterns, the Fallen Soldiers Memorial Day exhibits a completely different behavioral profile. In this way, the results exemplify an “affect rate” which reflects the significance and effect of specific times (two memorial days) on cultural production patterns (as measured by Instagram activity) in a specific place.


Image plot of 33,292 photos from Tel Aviv uploaded to Instagram between 20-26 April
A close-up of the visualization
A close-up of the visualization showing visually similar photos which document the air show during Independence Day morning
Figure 15: (a) Image plot of 33,292 photos from Tel Aviv uploaded to Instagram between 20–26 April 2012. The photos are organized by upload time (x–axis) and hue (y–axis) (b) A close–up of the visualization (c) A further close–up of the visualization showing visually similar photos which document the air show during Independence Day morning. A higher resolution version of this figure is available at:


3. In transition

Israel’s Independence Day celebrations begin directly after the end of the Fallen Soldiers Remembrance Day. This is an abrupt moment of transition in which the Israelis are asked to quickly switch from practicing memorial rituals to celebratory routines. We can see this drastic change in our results. While during regular days the number of photos uploaded every hour increases into late afternoon and then gradually decreases into the evening, Independence Day eve (25 April) exhibits a unique pattern: the number of photos continuously increases until 11pm. The cultural production rate continues to be significantly higher in later hours as people stay out later to celebrate. During Independence Day itself, there is a constant increase in cultural production until a peak around 2pm (Figure 14b).


Montage visualization of 33,292 photos taken in Tel Aviv form 20-26 April
A close-up of the visualization
A close-up of the visualization that captures users documenting the fireworks as part of Independence Day eve celebrations
Figure 16: (a) Montage visualization of 33,292 photos taken in Tel Aviv during April 20–26 2012, sorted by upload date (left to right, top to bottom). (b) A close–up of the visualization (c) A further close–up of the visualization that shows photographs of fireworks taken during Independence Day eve celebrations. A higher resolution version of this figure is available at:


4. Time–based affinities

As these results show, and as we will discuss hereunder, our imaginary communities (Instagram users situated within Tel Aviv) take different forms and shapes not only in terms of their aggregated dispersed or condensed spatial patterns, but also in their specific times (when the pictures were taken) and specific places (where they were taken). As opposed to many other maps of social media data that show social and spatial proximities in the form of aggregated location information from many moments and many people — thus producing singular maps where the different temporal origins of the data points are erased — we use visualization techniques which allow us to compare patterns between days, hours, locations and particular users, and see how the social status and function of a place change over “regular” and exceptional times. There are a number of such visualizations in this article. These visualizations show all locations data for every date over three months, colored by hour (Figure 11); Two weeks data shown as bar graphs indicating volume — as shown in the graphs above; or use the actual photos in a plot or montage, sorted by users (Figure 17) or various other visual attributes (Figures 1, 15, 16).


a) Montage visualization of 100 most active users in Tel Aviv area between 18-26 April
b) A close-up of the visualization
Figure 17: (a) Montage visualization of 100 Instagram users in Tel Aviv area who uploaded most photos during 18–26 April 2012. Each user’s photos appear in a single row sorted by upload date. (b) A close–up of the visualization (Visualizations were rotated by 90 degrees).


We can also combine some of these techniques into a singular visualization that will allow us to explore particular photos in specific locations and times. For example, Figure 18 uses image format and incorporates the location of photos (y axis), the time of creation (x axis) and the photos themselves, and allows us to explore photo–taking patterns in specific places in the city over time. As we can see, some places appear time and again as centers of concentration (i.e., Rabin square) while others perform as ad hoc cultural production centers on exceptional occasions.


Plot visualization of 33,292 photos from Tel Aviv between 20-26 April, sorted by time (x axis) and location (y axis)
Figure 18: Image plot visualization of 33,292 photos taken Tel Aviv during 20–26 April 2012, sorted by time (x axis) and location (y axis). Notice the significant changes in photo taking patterns around the city in exceptional vs. regular days, especially around Rabin square during Independence Day eve (increase in volume) and around Rothschild Blvd during Memorial Day eve (decrease in volume).


In a similar way, Figure 19 is a radial visualization that organizes the images according to their upload dates and locations. These new visualization forms combine the spatial, the temporal and the visual into a condensed representation. They allow us to better detect constantly changing sets of relations between Instagram photos across time, or during exceptional times and in specific places. They show how depending on the time of day, users tend to take pictures in different places, and how the nature of these places changes throughout the day and over longer time periods. We can then articulate these visualized relationships as “time–based affinities”: a set of relations between places or users at a specific point of time.


a) Radial image plot visualization of 7,328 photos taken in Tel Aviv, 25-26 April 2012
b) A close-up of the visualization
b) A close-up of the visualization
Figure 19: (a) Radial image plot visualization of 11,758 photos shared on Instagram in Tel Aviv during 25–26 April 2012. The photos are organized by date and time (angle) and location (radius). (b) (c) Close–ups of the visualization. The location position is obtained by multiplying latitude and longitude coordinates together. This allows us to visualize two spatial dimensions and the time dimension together in 2D plot. A higher resolution version of this figure is available at:


Complementary affinity

The most typical time–affinity that we identify in our visualizations can be called a complementary affinity: a set of relations between places that “complement” each other or relations between places that operate in a similar way during different times of the day. We then find groups of morning places, evening places, and so on, each group representing different characteristics and functions during various times of the day.

These fairly stable patterns also appear in relation to users’ affinities and the variety of ways in which we can typify and categorize people's behavior (Figure 20). As we can clearly see, a few users take many photos in one area, others move rapidly across the city. Several users never take more than one photo per hour, while others take many photos over short periods of time. Some take more photos during early mornings, while others only take photos during the late evening.


a) Matrix plot comparing activity of 289 most active Instagram users in Tel Aviv
b) A detail of the matrix plot showing eight users
Figure 20: (a) Matrix plot comparing activity of 289 most active Instagram users in Tel Aviv. Each plot in the matrix shows locations of photos shared on Instagram in Tel Aviv area over three months. The green to red color gradient indicates the time when a photograph was shared (green — morning, yellow — afternoon, red — evening). A line is drawn between two photos/dots that were taken within the same hour. (b) A detail of the matrix plot showing eight users. A higher resolution version of this figure is available at:


However, as illustrated in our study of Israeli memorial and independence days, this type of constant or stable network of affinities breaks during exceptional dates. During these times, we notice how the network of relations and connections changes, and the nature of this change can reflect the character of the time in which it occurs. During Independence Day eve, for example, we see many concentrations of small groups that gather in various places according to a similar interest in the specific nature of the celebration. This type of complementary affinity operates as a dispersed celebratory network that happens at the same time, but each node operates independently and exhibits different characteristics (Figure 21).


Radial plot visualization showing a subset of photos taken by Instagram users in Tel Aviv between 4pm 25 April and 2am 26 April 2012
Figure 21: Radial plot visualization showing a subset of photos taken by Instagram users in Tel Aviv between 4pm on 25 April and 2am on 26 April 2012. We used Amazon’s Mechanical Turk to separate photos that show people from photos with other subjects. This visualization includes only 2,268 photos with people (63 percent of all photos shared during this period). The photos are organized by location (angle) and upload date/time (radius). Location coordinates are obtained using the same method as figure 19. The visualization shows concentrations of photos in celebration locations around the city during Independence Day eve. Notice how the celebrations around Rothschild Blvd. begin significantly later and last longer than the celebrations around Rabin Square.


Memorial Day eve, however, exhibits a counter pattern to Independence Day eve activities. During these hours we see how our dispersed imaginary communities display similar low activity rate in picture–taking during the entire evening. In this case, the imaginary community of that day presents itself as a unified “unproductive” whole, organized around a singular cultural hub (Rabin Square) (see Figures 22, 23).


Scatter plots showing locations and photo-sharing times. Left: 24 April (Memorial Day eve). Right: 25 April (Independence Day eve)
Figure 22: Scatter plots showing locations and photo-sharing times. Left: 24 April (Memorial Day eve). Right: 25 April (Independence Day eve). The green to red gradient indicates the time (green — morning; yellow — noon; red — evening).


Contradictory affinity

This set of complementary relational affinities is complicated when examined both on the micro and the macro scales. On the micro scale, for example, during Soldiers’ Memorial Eve we can locate two places that operate in complete contradiction to each other and which represent, each in its own way, a different political affiliation. Two central ceremonial events are performed at the same time in Tel Aviv during that evening: a conservative memorial ceremony that is identified with more nationalist ideas (around Rabin Square area), and an alternative memorial ceremony that explicitly disassociates from the traditional, national one, and carries a different political affiliation (Hangar 11).

What type of “affinity” do these two events create? In our data, we can see how the national ceremony is depicted with a concentration of photos taken during the performance of the ceremony. On the other hand, within the spatial boundaries of the alternative ceremony almost no pictures were taken during that evening (Figure 23). While one place/event manifests a high cultural production rate as part of its memorial routine, the other place/event leads to a state of complete silence in remembering the dead [18]. We can call this a state of “contradictory affinity”, where two or more places are in a state of friction with each other (as opposed to the state of accordance we find on regular days).


In this scatter plot, we highlighted two memorial event locations during Memorial Day eve (24 April)
Figure 23: In this scatter plot, we highlighted two memorial event locations during Memorial Day eve (24 April). The same color gradient as in Figure 22 is used.


This same type of affinity also appears on the macro level, when we examine the most dramatic shift from Memorial Day to Independence Day. In less than 48 hours our visualizations exhibit two counter–representations that operate in complete opposition to each other (Figure 24). These contradictory representations show a 48–hour time span in which the routine patterns of earlier days are completely disrupted. These are representations of loud affinity (Independence Day) versus quiet affinity (Memorial Day) — each representing the changing nature of “city saturation” levels, indicating the rate and spread of images the city produces. In this way, during Memorial Eve many users choose not to take photos, and those who do, tend to concentrate around a unified location in the city. Independence Eve and Independence Day, however, show opposite patterns where larger numbers of users take photos, and a large proportion of these users take photos at various places at the same time of the day as well.


Scatter plot visualizations, with lines connecting the points (to highlight the difference in the patterns between the two days)
Figure 24: Scatter plot visualizations, with lines connecting the points (to highlight the difference in the patterns between the two days). The same color gradient as in Figures 22 and 23 is used. Left: Memorial Day eve (24 April). Right: Independence Day eve (25 April).


When this becomes clear, a more general operational principle emerges: as the level of affinity rises (quiet, loud, complementary, contradictory etc.), the city moves from a state of “normality” to “abnormality”. This latter state, as already mentioned, enables us to better visualize a cohesive whole. Indeed, as one might argue, the state of low or loud co–presence of Instagram users in the same space and in the same time does not necessarily create a “true” community. However, by tracing the changing nature of aggregated patterns in specific times and places, our visualizations produce imaginary communities that represent social realities in ways unavailable before.




This article integrates methods from social computing, digital humanities, and software studies to analyze visual social media. Our key research contributions are as follows: (1) We discuss Instagram platform as a software artifact, analyzing its interface, affordances and user experience. (2) Using large sets of Instagram photos for our case study, we show how visual social media can be analyzed at multiple spatial and temporal scales. (3) While most studies of social media focus on global patterns, we present analysis of social and cultural dynamics in specific places and particular times (national event days in Tel Aviv, Israel). (4) We introduce new visualization techniques which can show tens of thousands of individual images sorted by their metadata or algorithmically extracted visual features.

We began this article with an analysis of the Instagram interface, and the ways in which its affordances structure users’ particular cultural experiences. We then explored the ways in which we can visualize and analyze the visual content of social media data on a variety of scales. Starting at the global scale, we compared “visual signatures” of 13 global cities as they are represented in Instagram photos. Zooming into our data, we analyzed spatio–temporal patterns of over 200,000 Instagram photos uploaded in Tel Aviv, Israel over a three–month period. Finally, zooming further into the data, we focused on two weeks in Tel Aviv in order to show how temporal changes in numbers of shared photos, their locations, and visual characteristics can offer social, cultural and political insights about people’s activity during these dates.


Figure 25: Use of different filters in photos uploaded in Tel Aviv during April 2012. Each radial visualization includes photos with a particular filter. Radius: hue. Perimeter: upload time. Top left: X-Pro II. Top right: Lo-Fi. Bottom left: Amaro. Bottom right: Normal. Higher resolution versions of these visualizations are available at:


Affordances of Instagram’s software, as we have analyzed them, and our methods for exploring Instagram photos via multi–scale visualizations may signal a conceptual cultural shift in the ways we experience, analyze and use cultural data from the Internet. Recent cultural software tools and services (Instagram as well as our visualization tools) are less focused on organizing information and media into pre–existing structures and distinct categories. Instead, they enable the exploration of diversity of spatio–temporal and visual knowledge productions, and chart transitions and functions of particulars in relation to wholes (for example, exploring photos using tags, hashtags, locations, or follow particular users, as opposed to only using hierarchical subject categories). Imagine, for example, browsing through Instagram’s particular photos using its default application, then quickly visualizing millions of images from various locations, shifting constantly from the particular to the general, positioning oneself in multiple contexts and scales, moving from one location to another, all the while noting differences, similarities and intriguing relations and patterns.

While the most recent media transition known as Web 2.0 can be described in terms of a shift “from messages (made by other people) to platforms” where users can share, comment and tag their own media [19], we are now moving “from platforms to aggregators” that collect data streams from existing social information sources through API calls and organize it according to multiple attributes such as keywords, time, location, hashtags, etc. These aggregation systems present different data streams and act as “live stream readers” that pull together data from various social networks (known as ‘social network aggregators’ [20]), or as “analytics dashboards” that provide a synthesized and often algorithmically summarized views of data streams to extract meaningful insights (known as ‘social media control center’ [21]).

In this sense, our media visualizations participate in this media shift and illustrate a potential way in which collective social data activities turn into dynamic configurable patterns; they provide the ability to think of ways in which users can browse through non– (or less) hierarchical information based on intrinsic attributes (such as time, place, color, composition, presence or absences of faces, etc.) while re–arranging it in multiple contexts and scales (Figure 25).

If functions and relations are now more important than purposes, and we are, as previously suggested, encouraged to see ourselves as specific points of time and place, then we are also prompted to think of ourselves as singularities which are part of various wholes, each contributing to a constantly growing database that then needs to be visualized and explored. This is the essence of this new “media paradigm”: exploring diversities of singularities not through hierarchies and categories but rather through relations, transition and sequences, while moving from the singular to the plural, from the close to the distant. End of article


About the authors

Nadav Hochman is a doctoral candidate in the History of Art and Architecture Department at the University of Pittsburgh.
E–mail: nah61 [at] pitt [dot] edu

Lev Manovich is a Professor at The Graduate Center, CUNY, a Director of the Software Studies Initiative, and a Visiting Professor at European Graduate School (EGS). He is the author of Software Takes Command (Bloomsbury Academic, 2013), Soft Cinema: Navigating the Database (with Andreas Kratky, MIT Press, 2005), and The Language of New Media (MIT Press, 2001).
E–mail: manovich [dot] lev [at] gmail [dot] com



1. Originally the application was only available for mobile phones. As of February 2013 Instagram added a Web interface to allow people explore photos using Web browsers. In addition, on June 2013, Instagram added a new feature that allows users to shoot and share 15-seconds videos.

2. Official usage statistics around the world are not yet available. The most recent report mentions 130 million monthly active users, 16 billion shared photos, 40 million photos per day, 8500 likes per second, and 1000 comments per second. see:;;, accessed 21 June 2013.

3. Presner, forthcoming.

4. See for example: “Mapping the Republic of Letters, Stanford University,” at; “Cultures of Knowledge,” Oxford University, at; “Mapping Gothic France,” Columbia University, at; “HyperCities,” UCLA, at, accessed 19 February 2013; Currid and Williams, 2010, pp. 423–451.

5., accessed 2 November 2012.

6. Note that users can view their photos or isolate other users’ photos by launching their private pages within the application. However, this is not the default way of viewing the application. Compare with Flickr, the largest image and video hosting website to date, where photos are organized in personal photostreams with clear time indications. See: Flickr at, accessed 2 November 2012.

7. Note that you can choose not to use a filter on an image by applying a filter titled “normal”. In any case, the image still conforms to other manipulation tools such as size, lens etc.

8. This aligns well with what several writers identified as the new state of ‘atemporality.’ See for example: Sterling, 2010.

9. This cultural logic becomes even clearer if we consider other recent technological efforts that utilize a similar mechanism. Think for example of the Recapcha — an anti–spam technology in which users are required to decipher texts as part of a validation process and thus protect websites from automated programs written to generate spam. These texts are taken from digitized books and newspapers that optical character recognition (OCR) software has been unable to read. The deciphered results are then returned to the reCAPTCHA service, which sends them to the digitization projects. This new logic underlies various software that force users (even if not yet fully aware of) to actively participate and contribute to world knowledge. See:, accessed 2 November 2012.

10., accessed 2 November 2012.

11., accessed 2 November 2012.

12. “Google Earth now includes publicly–sourced aerial images from balloons and kites,” at, accessed 2 November 2012. Other software tools exhibit similar logic. An earlier prominent example is the service Mappr! (2005), a Web mashup service that combines a geographic map and photos from Flickr. See: Mappr! at, accessed 2 November 2012. For a more recent example, see also: Historypin at, accessed 2 November 2012.

13. Duggan and Brenner, 2013. Note that Pew Internet only surveyed users in the U.S., so we do not know exact proportions in other countries.

14. Lefebvre, 2004. See also Hägerstrand’s earlier work on time–geography (1975) that emphasized the time component in geographical representations and aimed to frame space and time together, without prioritizing one over the other.

15. Ibid., p.163.

16. Note that this conclusion only holds for general visual characteristics of photos like brightness, hue, saturation and texture. In future work we plan to add analysis of differences in photos’ content.

17. In addition to studying the differences between Instagram cities using visual features of the photos, we also compared the metadata for these photos. This analysis also shows that each city has its own character. For example, the proportions of “active users” (people who shared more than 30 photos during the period for which we collected data) varies significantly between the cities (Table 1).

18. It should be emphasized that there are significant differences in the number of people that attended each event. While the national ceremony attracts many thousands of people, the alternative ceremony is smaller and attracts just a few thousand people. In addition, while the national ceremony is preformed in an open square the alternative ceremony is conducted in an enclosed building. However, since we can see photos from the alternative ceremony location (Hangar 11) in earlier days, the lack of images on that date bears cultural significance.

19. Manovich, 2012.

20., accessed 22 June 2013.

21., accessed 22 June 2013.



Natalia Andrienko, Gennady Andrienko, Hendrik Stange, Thomas Liebig, and Dirk Hecker, 2012. “Visual analytics for understanding spatial situations from episodic movement data,” Kuünstliche Intelligenz, volume 26, number 3, pp.241–251.

Vyron Antoniou, Jeremy Morley, and Mordechai Haklay, 2010. “Web 2.0 geotagged photos: Assessing the spatial dimension of the phenomenon,” Geomatica, volume 64, number 1, pp. 99–110.

David J. Crandall, Lars Backstrom, Daniel Huttenlocher, and Jon Kleinberg, 2009. “Mapping the world’s photos,” WWW ’09: Proceedings of the 18th International Conference on World Wide Web, pp. 761–770.

Justin Cranshaw, Raz Schwartz, Jason I. Hong, and Norman Sadeh, 2012. “The Livehoods Project: Utilizing Social Media to Understand the Dynamics of A City,” Proceedings of the the Sixth International AAAI Conference on Weblogs and Social Media (ICWSM–12), Dublin, Ireland, pp. 1–8, and at, accessed 16 June 2013.

Elizabeth Currid and Williams Sarah, 2010. “The geography of buzz: Art, culture and the social milieu in Los Angeles and New York,” Journal of Economic Geography, volume 10, number 3, pp. 423–451.

Peter S. Dodds, Kameron D. Harris, Isabel M. Kloumann, Catherine A. Bliss, and Christopher M. Danforth, 2011. “Temporal patterns of happiness and information in a global social network: Hedonometrics and Twitter,” PLOS ONE, volume 6, number 12, at, accessed 26 February 2013.

Carl Doersch, Saurabh Singh, Abhinav Gupta, Josef Sivic, and Alexei A. Efros. 2012. “What makes Paris look like Paris?” ACM Transactions on Graphics (SIGGRAPH 2012 Conference Proceedings), volume 31, number 4, article number 101.

Maeve Duggan and Joanna Brenner, 2013. “The demographics of social media users — 2012,” Pew Internet & American Life Project (14 February), at, accessed 6 March 2013.

Franck Ernewein, 2013. “Tweetping,” at, accessed 5 March 2013.

Eric Fischer, 2010. “The geotaggers’ world atlas,” at, accessed 5 March 2013.

Nir Grinberg, Mor Naaman, Blake Shaw, and Gilad Lotan, “Extracting diurnal patterns of real world activity from social media,” Proceedings of the Seventh International AAAI Conference on Weblogs and Social Media (ICWSM–13), and at, accessed 16 June 2013.

Torsten Hägerstrand, 1975. “Space, time and human conditions,” In: Anders Karlqvist, Lars Lundqvist, and Folke Snickars (editors). Dynamic allocation of urban space. Lexington, Mass.: Lexington Books. pp. 3–14.

James Hays and Alexei A. Efros, 2008. “IM2GPS: Estimating geographic information from a single image,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1–8, and at, accessed 16 June 2013.

Nadav Hochman and Raz Schwartz, 2012. “Visualizing Instagram: Tracing cultural visual rhythms,” Proceedings of the Workshop on Social Media Visualization (SocMedVis) in conjunction with the Sixth International AAAI Conference on Weblogs and Social Media (ICWSM–12), pp. 6–9, and at, accessed 16 June 2013.

Alexander Jaffe, Mor Naaman, Tamir Tassa, and Marc Davis, 2006. “Generating summaries and visualization for large collections of geo–referenced photographs,” MIR ’06: Proceedings of the 8th ACM International Workshop on Multimedia Information Retrieval, pp. 89–98.

Lyndon S. Kennedy and Mor Naaman, 2008. “Generating diverse and representative image search results for landmarks,” WWW ’08: Proceedings of the 17th international Conference on World Wide Web, pp. 297–306.

Slava Kisilevich, Milos Krstajic, Daniel Keim, Natalia Andrienko, and Gennady Andrienko, 2010. “Event–based analysis of people’s activities and behavior using Flickr and Panoramio geotagged photo collections,” IV ’10: Proceedings of the 2010 14th International Conference Information Visualisation, pp. 289–296.

Kalev H. Leetaru, 2012. “SGI global Twitter heartbeat,” at, accessed 26 February 2013.

Henri Lefebvre, 2004. Rhythmanalysis: Space, time, and everyday life. Translated by Stuart Elden and Gerald Moore. New York: Continuum.

Xirong Li, Cees G.M. Snoek, and Marcel Worring, 2009. “Annotating images by harnessing worldwide user–tagged photos,” ICASSP ’09: Proceedings of the 2009 IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 3,717–3,720.

Yunpeng Li, David J. Crandall, and Daniel P. Huttenlocher, 2009. “Landmark classification in large–scale image collections,” Proceedings of the 2009 IEEE 12th International Conference on Computer Vision, pp. 1,957–1,964; version at, accessed 17 June 2013.

Lev Manovich, 2013. Software takes command. New York: Bloomsbury Academic.

Lev Manovich, 2012. “How to follow software users? (Digital humanities, software studies, big data),” at, accessed 2 November 2012.

Mor Naaman, Amy X. Zhang, Samuel Brody, and Gilad Lotan, 2012. “On the study of diurnal urban routines on Twitter,” Proceedings of the Sixth International AAAI Conference on Weblogs and Social Media (ICWSM–12), at, accessed 17 June 2013.

Todd Presner, forthcoming. HyperCities: Thick mapping in the digital humanities. Cambridge, Mass.: Harvard University Press.

Jose San Pedro and Stefan Siersdorfer, 2009. “Ranking and classifying attractiveness of photos in folksonomies,” WWW ’09: Proceedings of the 18th International Conference on the World Wide Web, pp. 771–780.

Ian Simon, Noah Snavely, and Steven M. Seitz, 2007. “Scene summarization for online image collections,” ICCV ’07: Proceedings of the 11th IEEE international Conference on Computer Vision, pp. 1–8; version at, accessed 17 June 2013.

Bruce Sterling, 2010. “Atemporality for the creative artist,” Wired (25 February), at, accessed 2 November 2012.

Katerina Vrotsou, Natalia Andrienko, Gennady Andrienko, and Piotr Jankowski, 2011. “Exploring city structure from georeferenced photos using graph centrality measures,” In: Dimitrios Gunopulos, Thomas Hofmann, Donato Malerba, and Michalis Vazirgiannis (editors). Machine learning and knowledge discovery in databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5–9, 2011, Proceedings, Part III. Lecture Notes in Computer Science, volume 6913, pp. 654–657.

Yu Zheng and Jason Hong, 2012. “The preface of the 4th International Workshop on Location–Based Social Networks,” UbiComp ’12: Proceedings of the 2012 ACM Conference on Ubiquitous Computing, pp. 894–896, and at, accessed 17 June 2013.


Editorial history

Received 8 May 2013; accepted 3 June 2013.

Copyright © 2013, First Monday.
Copyright © 2013, Nadav Hochman and Lev Manovich.

Zooming into an Instagram City: Reading the local through social media
by Nadav Hochman and Lev Manovich.
First Monday, Volume 18, Number 7 - 1 July 2013